On Wed, Jan 19, 2011 at 12:10 PM, Andrew S?guin <aseguo at gmail.com> wrote: > Hi, > > Assuming the replication works fine (you see the files on both servers > A and B) and that you use the Gluster client (mount -t glusterfs ...), > then if server A goes down, your application will continue seamlessly > to use server B although you could encounter problems later: > > If the server with your application has to restart, and server A is > down at that moment, then unfortunately the gluster client does not > connect to a gluster server, and does not get to know that server B > exists. > > To avoid that problem, you can prepare your Gluster cluster system > with 3 IP addresses: server A, server B, and a "shared" IP for the two > nodes which will be active only on one of the two servers at a time. > Some system then has to decide which server is currently master -- the > CARP protocol could be of help here. Maybe the following blog/howto > would be of help for this: > http://www.misdivision.com/blog/setting-up-a-highly-available-storage-cluster-using-glusterfs-and-ucarp > > With NFS & CIFS the clients don't know about the second server, in > case you do not have a shared IP, the connection would be broken. NFS > without the shared IP is the same as having an offline backup but with > the advantage that the backup is up to date. > > While I'm not a gluster expert, I hope this helps! > Andrew > > At this stage you need to use pNFS (parallel NFS) to make NFS more redundant :) -- Kind Regards Rudi Ahlers SoftDux Website: http://www.SoftDux.com Technical Blog: http://Blog.SoftDux.com Office: 087 805 9573 Cell: 082 554 7532
